diff options
| author | Jingyue Wu <jingyue@google.com> | 2015-07-31 21:44:14 +0000 |
|---|---|---|
| committer | Jingyue Wu <jingyue@google.com> | 2015-07-31 21:44:14 +0000 |
| commit | bd11ccb2b9c5ff62a37893e47d338d1620f7ba08 (patch) | |
| tree | 1bd01291c4cc1ee914776048fc1b7f587e751ce0 /test/CodeGen/NVPTX | |
| parent | be031d9158aa3fbc87a5f9e2c1b4b74a414cd0b6 (diff) | |
[NVPTX] convert pointers in byval kernel arguments to global
Summary:
For example, in
struct S {
int *x;
int *y;
};
__global__ void foo(S s) {
int *b = s.y;
// use b
}
"b" is guaranteed to point to global. NVPTX should emit ld.global/st.global for
accessing "b".
Reviewers: jholewinski
Subscribers: llvm-commits, jholewinski
Differential Revision: http://reviews.llvm.org/D11505
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243790 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/NVPTX')
| -rw-r--r-- | test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll | 20 |
1 files changed, 18 insertions, 2 deletions
diff --git a/test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll b/test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll index 0de72c4a1ae..2fffa3eeac1 100644 --- a/test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll +++ b/test/CodeGen/NVPTX/lower-kernel-ptr-arg.ll @@ -1,7 +1,7 @@ ; RUN: llc < %s -march=nvptx64 -mcpu=sm_20 | FileCheck %s target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64" -target triple = "nvptx64-unknown-unknown" +target triple = "nvptx64-nvidia-cuda" ; Verify that both %input and %output are converted to global pointers and then ; addrspacecast'ed back to the original type. @@ -26,6 +26,22 @@ define void @kernel2(float addrspace(1)* %input, float addrspace(1)* %output) { ret void } -!nvvm.annotations = !{!0, !1} +%struct.S = type { i32*, i32* } + +define void @ptr_in_byval(%struct.S* byval %input, i32* %output) { +; CHECK-LABEL: .visible .entry ptr_in_byval( +; CHECK: cvta.to.global.u64 +; CHECK: cvta.to.global.u64 + %b_ptr = getelementptr inbounds %struct.S, %struct.S* %input, i64 0, i32 1 + %b = load i32*, i32** %b_ptr, align 4 + %v = load i32, i32* %b, align 4 +; CHECK: ld.global.u32 + store i32 %v, i32* %output, align 4 +; CHECK: st.global.u32 + ret void +} + +!nvvm.annotations = !{!0, !1, !2} !0 = !{void (float*, float*)* @kernel, !"kernel", i32 1} !1 = !{void (float addrspace(1)*, float addrspace(1)*)* @kernel2, !"kernel", i32 1} +!2 = !{void (%struct.S*, i32*)* @ptr_in_byval, !"kernel", i32 1} |